Welcome to AdversarialTensors's documentation!#

Overview#

This library implements a variety of tensor factorization methods for defending Artificial Intelligence (AI) models against adversarial attacks.

The library provides the following main functionalities:

  1. Preprocessing stage that uses tensor factorization methods to reduce adversarial noise.

  2. Identifying novel latent attack features by combining different attack methods.

  3. Employing Generative Adversarial Networks (GANs) to generate denoised data.

Capabilities#

  • Tensor factorizations methods like Tucker, non-negative Tucker, CPD, non-negative CPD, Tensor Train, NMF, NNSVD and NMF.

_images/denoising.png

  • Ability to denoise image in various modes: single, batch, and all.

_images/inference.png

  • Attacks such as AutoAttack, FGSM, PGD, BIM, DeepFool.

  • White-box and black-box attack capabilities.

_images/white_box.png _images/black_box.png

  • Ability to perform adversarial training.

_images/white_box_adv_train.png _images/black_box_adv_train.png

  • Support for datasets: Cifar10, Cifar100, Imagenet, Tiny-Imagenet, MNIST, Fashion-MNIST.

End-to-End Code Execution Guide#

Training Models#

First, set the environment variables specific to your experiment in send_train_jobs.sh.

Output:

  • Models will be stored in my_checkpoints. For example: my_checkpoints/cifar10/resnet18_0_1.ckpt

  • If using CSVLogger, log data will be saved in log_{dataset_name}_{model_name}_test/{fold_ind}_{nfolds}

To Run:

bash send_train_jobs.sh

Generating Attack Data#

Set the environment variables for the experiment in send_attack_jobs.sh.

Output:

  • Attack data and logs will be stored in attack_log_{dataset_name}/{model_name}/{fold_ind}_{nfolds}_{attack_type}

To Run:

bash send_attack_jobs.sh

Parameter Tuning#

Set the environment variables for the experiment in send_tune_jobs.sh.

Output:

  • The tuning results will be located at parameter_search_{dataset_name}/{model_name}/{fold_ind}_{nfolds}_{attack_type}/search_results

To Run:

bash send_tune_jobs.sh

Collecting Results#

After setting the environment variables:

Output:

  • The results will be stored as CSV files in tune_csv_results

To Run:

bash collect_results.sh

Final Evaluation on Test Set#

Generating Attacks

  • Set final_test_set to True in send_attack_jobs.sh

bash send_attack_jobs.sh
Evaluating Models

  • After setting specific environment variables (such as eval_top_k):

To Run:

bash send_final_eval_jobs.sh

Indices and tables#